Some Practical Advice On That HMC-Power9 Impedance Mismatch
March 1, 2021 Timothy Prickett Morgan
In any modern IT infrastructure – be it compute, storage, or networking – there is an increasing architectural movement to break control planes from the compute, storage, or networking planes. In this sense, the Hardware Management Console, which people have been complaining about since it was launched so long ago we can’t even remember it.
The HMC debuted as an external controller for system configuration and logical partition configuration with the Power5-based “Squadron” line of servers running OS/400 back in 2004, including the Power 520 and the Power 570 as well as the Power 575, Power 590, and Power 595 machines, after they had been in use on pSeries and zSeries systems for years before then. The HMC costs on the order of $3,500 to $4,000, which is a pain in the neck and which added another layer of complexity to iSeries, System i, and IBM i platforms. But again, it is absolutely normal to want to separate the control plane for system management from the data and application planes – all the hyperscalers do this with all of their platforms. The reason is that by separating the control plane from the data/application plane, the configuration of the system is decoupled from the running of that system and any issues with that system will not and cannot impact the control plane. This is also inherently more secure because that control plane is not in the data path or application path. Conversely, if the control plane fails, the system can keep running as currently configured and it does not crash.
Given all of this, it is a wonder that all IBM i machines do not require at least a virtualized HMC. But, the OS/400 and IBM i base has its habits and is, like every other established customer base, resistant to change. So the HMC has been adopted only where absolutely necessary. To be fair, IBM did not make it affordable in the early years of HMC deployment, and only in recent years has the HMC been available as a virtual appliance running atop the PowerVM hypervisor. But if customers do that, then they are introducing a potential single point of failure on the system since PowerVM can crash and take everything down. (The odds don’t favor this, of course, but it is not good architecture to embed the control plane in the same virtual machine layer it is controlling.)
The HMC has been running on Linux on an X86 rack server since it was launched, but in August 2017, IBM ported the HMC code to run natively on a Power-based console and also on a virtual machine running on top of the PowerVM server virtualization hypervisor. This is all good. But the killing off of support for managing Power9 servers with a certain firmware level on X86-based HMC hardware is causing some problems.
A few weeks ago, IBM updated its announcement about this, clarifying that the FW950 firmware level announced for updated Power9 machines back on November 23, 2020 would require a minimum HMC software level of V9R2M950. However, the X86-based HMCs – including machines with model numbers 7042-CR9, 7042-CR8, 7042-CR7 – cannot support this software level and are stuck at V9R1 software. So they therefore cannot be used to manage the updated Power9 systems.
This means customers are going to have to either buy a new 7063-CR1 Power-based HMC or run it in a Virtual HMC V9 appliance if they want to manage these newer Power9 machines or older ones that have their firmware updated to the FW950 level.
A lot of companies run their monitoring and control plane software out in a secure cloud, not only distinct from their systems, but absolutely separate from them. We wonder why IBM doesn’t offer the HMC as a service as just eliminate this problem entirely. In the age of COVID-19, working remotely is the rule, not the exception, and moreover, companies don’t want to have to pay for an HMC or manage it. They just want a secure tool that allows them to manage their Power9 machines and their IBM i, AIX, and Linux partitions. No one wanted an HMC in the first place, and some days – remember when IBM tried to replace the HMC with the Systems Director Management Console a decade ago? – we think IBM doesn’t either. This would solve that problem, once and for all and also get all customers using the HMC if it just came with the system as a service, like Electronic Customer Support did with the original AS/400s.
It’s something to think about.
RELATED STORIES
IBM Moves HMC Management To Native Power Server, LPARs
Big Blue Rolls Out New HMC For Power Control Freakage
The SDMC Is Dead, Long Live The HMC
IBM Gives The 7042-CR6 HMC Another Execution Stay
More Reader Feedback On Control Your Code, And HMC Stay Of Execution
IBM Launches New HMC for Power 5 Servers
Some Insight Into the HMC-to-SDMC Transition
IBM Replaces the HMC with New Systems Director Console
IBM Delivers Deskside HMC Console for PowerVM Management
Internet Explorer 8 Doesn’t Support HMC V7. . . or Does It?
IBM Launches Upgraded, Rack-Mounted HMC for Partition Control
Trouble-Shooting WebSM to HMC Connectivity Problems
Admin Alert: Remotely Accessing an HMC System Console, Part 2
Admin Alert: Remotely Accessing an HMC System Console, Part 1
IBM Previews Virtualization Management Tool for Power-Based Boxes
IBM Announces New HMCs for System p and System i Servers
IBM Launches New HMC for Power 5 Servers
Sundry Summer Announcements for the System i5
I’m surprised that an OpenHMC initiative has not sprung up. I’m sure there are knowledgeable people out there would could easily reverse engineer the protocols used by the HMC, to at least be able to do the fundamental tasks like define/stop/start/dlpar and opening a virtual terminal.
Its getting to the point where you need a more compute just to manage the PowerVM environments – HMC + VIO servers, PowerVC and now PowerSC on top – abstraction layer on top of abstraction layer.
One of the advantages of PowerVM is that the hypervisor has a relatively low resource overhead, but if these additional components are now required, it makes the value proposition in comparison to VMWare a lot less appealing.
Running the HMC on the server that its managing doesn’t make much sense though – even in an environment with clustered servers, its the HMC that manages the DLPAR and remote restart – if the server running the HMC fails… Basically it means you need to run 2 HMCs, which is fine, but not when they are chewing vCPU and RAM on a new Power9 server, where the CPU and RAM resources cost a premium.
For IBM i p05 systems, that are already limited to 4 cores 64GB, its not an option to run a HMC on top of that, so you have to add a HMC, the alternative being to opt for a p10 class, and then you have to be very vigilant to ensure you don’t get charged for IBM i SWM on cores not allocated to it – I’ve seen customers charged IBM i software licensing for all cores in the box, despite the fact that the IBM i partitions were limited to 2 cores on the system.
VIO’s are already consuming considerable resources, adding a HMC on top may be the straw that breaks the camels back.